Z-Image De-Turbo De-distilled Model

De-distilled version of Z-Image model, breaking turbo distillation limits and restoring trainability and flexibility

De-distilled

LoRA Training

Deep Fine-tuning

ComfyUI

Trainability

Overview

Z-Image De-Turbo is a de-distilled version of Tongyi-MAI/Z-Image-Turbo, fine-tuned on images generated by Z-Image-Turbo to break down the turbo distillation limitations. This model is specifically designed for training and deep fine-tuning, offering enhanced trainability and flexibility compared to the original turbo model.

Features

De-distillation technology breaking original turbo model limitations
Direct training capability without adapters
LoRA training support while maintaining compatibility with base model
Deep fine-tuning support exceeding original turbo model capabilities
ComfyUI version and diffusers-based version available
CFG normalization support

Images

Sample image generated using Z-Image De-Turbo

Installation

git clone https://huggingface.co/ostris/Z-Image-De-Turbo
pip install -r requirements.txt

Usage

For inference, use low CFG (2.0-3.0) and 20-30 steps. The model supports CFG normalization for better generation results.

Requirements

Python 3.8+
PyTorch
Diffusers library
CUDA compatible GPU
16GB+ VRAM recommended for training